Calos Fund Limited Liability Company

Ol Security Limited Liability Company

Nvidia Corporation

Stream Processors, Inc.

Ol Security Limited Liability

Ujval J Kapasi

Yipeng Liu

Amit Gulati

Dan Miller

Brucek Kurdo Khailany

William James Dally

Donald James Curry

John C Sievers

Jim Jian Lin

John Seivers

Hao Fang Wu

Yilin Zhang

Dilip Sequeira

Alex Fit-Florea

Slawomir Kierat

Maxim Milakov

John Sievers

Szymon Migacz

Zacky Zhou

Instructions embedded on a computer-readable medium, when executed on one or more computer devices, improve video coding performance while using a merge mode in motion estimation. The instructions comprise instructions to perform one or more refinement searches on a plurality of candidate regions of a current frame. The instructions also comprise instructions to determine one or more distortion values based, at least in part, on reduced candidate regions and instructions to code motion data based, at least in part, on the one or more refinement searches.

Hierarchical packing of syntax elements

A method operates within an integrated circuit having a plurality of processing lanes. For each of a first and second processing lanes, the method determines a number of packed data words among one or more packed data words associated with the respective processing lane, associates the number of packed data words with a used field of the processing lane, wherein the used field indicates the number of packed data words in the processing lane; and stores the one or more packed data words in a variable record length memory based, at least in part, on the used field of the processing lane.

Aspects of the present invention are directed to computer-implemented techniques for performing data compression and conversion between data formats of varying degrees of precision, and more particularly for improving the inferencing (application) of artificial neural networks using a reduced precision (e.g., INT8) data format. Embodiments of the present invention generate candidate conversions of data output, then employ a relative measure of quality to identify the candidate conversion with the greatest accuracy (i.e., least divergence from the original higher precision values). The representation can be then be used during inference to perform computations on the resulting output data.

Automated methods for conversions to a lower precision data format

A method operates within an integrated circuit device having a plurality of processing lanes. The method determines a first number of packs among one or more first packs associated with a first processing lane of the plurality of processing lanes, associates the first number of packs with a first used field of the first processing lane, determines a second number of packs among one or more second packs associated with a second processing lane of the plurality of processing lanes, and associates the second number of packs with a second used field of the second processing lane.

A method of operation within an integrated circuit device having a plurality of processing lanes. A first sub-stream of data, having a variable length, is generated in a first one of the processing lanes. A second sub-stream of data, also having a variable length, is generated in a second one of the processing lanes. The first and second sub-streams are then output to form a single bitstream.

In a single-instruction-multiple-data (SIMD) processor having multiple lanes, and local memory dedicated to each lane, a method of processing an image is disclosed. The method comprises mapping consecutive rasters of the image to consecutive lanes such that groups of consecutive rasters form image strips, and vertical stacks of strips comprise strip columns. Local memory allocates memory to the image strips. A sequence of functions is processed for execution on the SIMD processor in a pipeline implementation, such that the pipeline loops over portions of the image in multiple iterations, and intermediate data processed during the functions is stored in the local memory. Data associated with the image is traversed by first processing image strips from top to bottom in a left-most strip column, then progressing to each adjacent unprocessed strip column.

Chaining image-processing functions on a SIMD processor

Method for chaining image-processing functions on a SIMD processor

Disclosed are methods and systems for dynamically determining data-transfer paths. The data-transfer paths are dynamically determined in response to an instruction that facilitates data transfer among execution lanes in an integrated-circuit processing device operable to execute operations in parallel. In addition, embodiments include an integrated-circuit processing device operable to execute operations in parallel, including the capability of providing confirmation information to potential source lanes, the confirmation information indicating whether the potential source lanes may send data to requested destination lanes during a data-transfer interval.

Data exchange and communication between execution units in a parallel processor

A method of estimating motion is disclosed. A first plurality of candidates is identified in a reference frame, wherein the total area occupied by the first plurality of candidates is substantially smaller than that of the reference frame. A first refinement search is then performed based, at least in part, on the first plurality of candidates. One or more best candidates are then identified based, at least in part, on the first refinement search. Finally, motion data is encoded based, at least in part, on the one or more best candidates.

Video coding on parallel processing systems

A method of operation within an integrated-circuit processing device having a plurality of execution lanes. Upon receiving an instruction to exchange data between the execution lanes, respective requests from the execution lanes are examined to determine a set of the execution lanes that may send data to one or more others of the execution lanes during a first interval. Each execution lane within the set of the execution lanes is signaled to indicate that the execution lane may send data to the one or others of the execution lanes.

A processor having a zero-overhead operand copy capability. The processor includes multiple execution units to execute instructions in parallel and multiple register files each associated with one or more of the execution units. The processor further includes circuitry to select either an instruction execution result from a first one of the execution units or content of a register within a first one of the register files associated with the first one of the execution units to be stored within a register within a second one of the register files.

Growing community of inventors

Westford, MA, United States of America

Ujval J Kapasi